I’m in the midst of reading Lewis Hyde’s Trickster Makes the World. It’s great, and I want to see whether the themes it identifies can be substantiated by existing folklore motifs and tale types.

Transforming the data

Get trickster motifs

First, we can try a simple query using the word “trickster”:

trickster <- 
  tmi %>% 
  filter(str_detect(motif_name,regex("trickster", ignore_case = T))) %>%
  select(id:level)

Next, let’s expand to look at any motif involving tricks. We’re going to use regular expressions (or, regex) to identify patterns in text, and we’ll save these as a list object:

regexes <- list()
regexes$trick <- "^trick | trick | trick$|^tricks | tricks | tricks$"

tricks <- 
  tmi %>% 
  filter(str_detect(motif_name,regex(regexes$trick,ignore_case = T)))

rm(trickster)

In Hyde’s book, he identifies a number of other themes which accompany trickster tales, such as:

  • Bait. Synonyms: bait, trap, snare, catch, net
  • Stealing. Synonyms: steal, rob
  • Transformation. Synonyms: shape-shift, transform
  • Boundaries. Synonyms: border, boundary, cross-roads

We can make some regexes as search strategies, but keep in mind that these should be refined by looking at the data that’s returned:

regexes$bait <- "bait|trap|snare|catch| net "
regexes$steal <- "steal|rob"
regexes$transform <- "shape-shift|in.*shape|take.*shape|assume.*shape|transform"
regexes$border <- "border|boundary|boundaries|cross-roads"

We can add to these anytime (after all, I haven’t even finished the book yet), but let’s start here and develop a subset of the tmi which contains trickster-related motifs.

tricks <-
  tmi %>%
  select(id:chapter_id) %>%
  mutate(
    trickster = if_else(str_detect(motif_name,regex("trickster", ignore_case = T)),T,F),
    trick = if_else(str_detect(motif_name,regex(regexes$trick,ignore_case = T)),T,F),
    bait = if_else(str_detect(motif_name,regex(regexes$bait,ignore_case = T)),T,F),
    steal = if_else(str_detect(motif_name,regex(regexes$steal,ignore_case = T)),T,F),
    transform = if_else(str_detect(motif_name,regex(regexes$transform,ignore_case = T)),T,F),
    border = if_else(str_detect(motif_name,regex(regexes$border,ignore_case = T)),T,F)
  ) %>%
  rowwise() %>%
  mutate(any = sum(c_across(trickster:border))) %>%
  ungroup() %>%
  filter(any > 0)

Exploratory Analysis

Common motifs

What are the most common motifs in tales that include trickster themes?

trickster_tales %>%
  group_by(motif) %>%
  summarize(n = n_distinct(atu_id)) %>%
  left_join(tmi %>% select(id,motif_name),by = c("motif" = "id")) %>%
  select(motif,motif_name,n) %>%
  slice_max(order_by = n, n = 15,with_ties = F) %>%
  rmarkdown::paged_table()

What if we group by the sections in which these motifs occur?

trickster_tales %>%
  group_by(level_1) %>%
  summarize(n_tales = n_distinct(atu_id)) %>%
  left_join(
    tmi %>% 
      filter(level == 1) %>%
      select(level_1,motif_name),
    by = c("level_1")
  ) %>%
  select(level_1,motif_name,n_tales) %>%
  slice_max(order_by = n_tales, n = 30,with_ties = F) %>%
  rmarkdown::paged_table()

Observations:

  • Some of these are just common motifs (e.g. L160, H1370)
  • The ‘trickster’ behavior may not involve a deity, so we’ll need to subset our original dataset to get better results

Co-occurence

To look at all motif co-occurrences, we can make a network:

library(tidygraph); library(visNetwork)

nodes <-
  trickster_tales %>%
  distinct(level_1) %>%
  left_join(
    tmi %>% 
      filter(level == 1) %>%
      select(level_1,motif_name),
    by = c("level_1")
  ) %>%
  mutate(id = row_number()) %>%
  select(id, label = level_1, title = motif_name)
  
edges <- 
  trickster_tales %>%
  select(atu_id,from = level_1) %>%
  group_by(atu_id) %>%
  mutate(to = lead(from, default = from[1])) %>%
  filter(from != to) %>%
  group_by(from,to) %>%
  summarise(n = n_distinct(atu_id)) %>%
  ungroup() %>%
  left_join(nodes %>% select(id,label), by = c("from" = "label")) %>%
  select(-from) %>%
  rename(from = id) %>%
  left_join(nodes %>% select(id,label), by = c("to" = "label")) %>%
  select(-to) %>%
  rename(to = id) %>%
  select(from,to,width = n)

graph <- 
  tbl_graph(nodes,edges) %>%
  activate(nodes) %>%
  mutate(
    btwn = centrality_betweenness(),
    eigen = centrality_eigen(),
    group = group_edge_betweenness(),
    value = scales::rescale(btwn, to = c(0,10))
  ) %>%
  activate(edges) %>%
  mutate(
    e_btwn = centrality_edge_betweenness(),
    width = scales::rescale(e_btwn, to = c(0.5,6))
  )

rm(nodes); rm(edges)

And graph it interactively:

library(visNetwork)
network <- toVisNetworkData(graph)
visNetwork(nodes = network$nodes, edges = network$edges) %>%
  visEdges(arrows = 'middle') %>%
  visIgraphLayout(layout = "layout_with_fr")

Some variations:

  • choose a different layout from igraph
  • apply different measures to the value to resize nodes, to the width argument to resize edges, or to the group argument to change the colors
  • additional resources can be found in the tidygraph package

We can also inspect the node and edge measures in a table:

graph %>% activate(nodes) %>% as_tibble() %>% View()

Concepts

We can also explore more abstract concepts present within the tales by mapping their language to standardized lexicons. First, we need to unpack (a.k.a. tokenize) the data so it has one word per row. This time, we’ll use the longer tale descriptions from the atu_df:

library(tidytext)

tale_words <-
  atu_df %>%
  filter(atu_id %in% tale_list) %>%
  select(atu_id, tale_type) %>%
  unnest_tokens(word, tale_type)

Now we’ll need to get some lexicons and join them to our data. We’ll start with:

  • The NRC Valence, Arousal, and Dominance (VAD) Lexicon. A lexicon rating the valence (i.e. positive/pleasure – negative/displeasure), arousal (i.e. active/stimulated – sluggish/bored), and dominance (i.e. powerful/strong – powerless/weak) of ~20k words. See link.
  • The Personal Values Dictionary. A lexicon which can be used to assess references to value orientations in textual data, such as tweets, blog posts, or status updates. Based on the work of Shalom Schwartz. See link.
library(textdata)
vad <- lexicon_nrc_vad()
# eil <- lexicon_nrc_eil()

values <- 
  read_delim("value_lexicon.txt","\t",skip = 12, col_names = c("word","code")) %>%
  left_join(
    read_delim("value_lexicon.txt","\t",skip = 1,n_max = 10,col_names = c("code","value")),
    by = "code"
  ) %>%
  mutate(
    value = str_remove(value,"^Values_"),
    value_desc = case_match(
      value,
      "AC" ~ "achievement", 
      "BE" ~ "benevolence",
      "CO" ~ "conformity",
      "HE" ~ "hedonism",
      "PO" ~ "power",
      "SD" ~ "self-direction",
      "SE" ~ "security",
      "ST" ~ "stimulation",
      "TR" ~ "tradition",
      "UN" ~ "universalism"
    )
  ) %>%
  select(-code)

tale_words <-
  tale_words %>%
  left_join(vad, by = c("word" = "Word")) %>%
  left_join(values)

We can compare the average valence (pos/neg), arousal (active/bored) and dominance (strong/weak) of the tales’ language:

p <- 
  tale_words %>%
  group_by(atu_id) %>%
  summarise(
    avg_valence = mean(Valence, na.rm = T),
    avg_arousal = mean(Arousal, na.rm = T),
    avg_dominance = mean(Dominance, na.rm = T)
  ) %>%
  left_join(atu_df %>% select(atu_id,tale_name)) %>%
  mutate(label = paste0(atu_id,": ", tale_name)) %>%
  ungroup() %>%
  ggplot(
    aes(x = avg_valence, y = avg_arousal, color = avg_dominance, label = label)
  ) +
  geom_point() +
  scale_color_viridis_c() +
  theme_minimal()

ggplotly(p)

Or we can check to see which values are highlighted in trickster tales, since these tales are considered by Hyde to be amoral and we might thus expect to see an even dispersion:

p <-
  tale_words %>%
  group_by(value_desc) %>%
  summarise(n = n()) %>%
  ungroup() %>%
  mutate(pct = round(n/sum(n) * 100, digits = 1)) %>%
  filter(!is.na(value_desc)) %>%
  ggplot(aes(x = value_desc, y = pct)) +
  geom_col() +
  theme_minimal()

ggplotly(p)

Observations:

  • It makes sense that universalism is low, since tricksters are known to confuse clear polarities/systems.
  • It’s unexpected (at least to me) to see benevolence score so high, as well as tradition, when we might expect trickster tales to be more balanced in these areas. May want to dive in and see where this occurs.
  • We might also want to know whether the proportion of these values in trickster tales are similar to those in the atu as a whole. This graph just shows us their relationship within the subset of tales we filtered for.

Resources

A few additional resources for text analysis in R: